Purpose
This study defines a FLACS biometric system repeatability measure based on the probability of image processing errors having an influence on the construction of a 3D lens model and subsequent laser treatment, and uses statistical analysis to evaluate repeatability as a function of the number of biometric images acquired and the automatic image processing performance.
Methods
Using MATLAB, a model was constructed to assess the impact of image processing errors on laser treatment. The model consisted of a biometric scan yielding N images, which were automatically processed by algorithms that either successfully processed, erroneously processed, or left unprocessed each image. The probability of each outcome is specified as an input parameter to the model, along with N. If N > 2, the model includes the use of outlier removal techniques when combining the images into the 3D lens model. With outlier removal, if there are fewer erroneously processed images than correctly processed images, the 3D lens model will be constructed as though the erroneously processed images had been left unprocessed, thus removing their influence on the treatment. The model uses established probability laws to compute the biometric system error rate (the probability of image processing errors influencing treatment) for the given input parameters. Biometric system repeatability is defined as the inverse of the error rate. Thus, an error rate of 1% yields a repeatability of 100, meaning one would expect to observe an error about once every 100 treatments.
Results
The MATLAB program was run on several different parameter sets. For all parameter sets tested, an exponential relationship was observed between the repeatability and the number of images acquired in the biometric scan (R squared values greater than 0.99). The parameter sets included per-image error rates (PIER) ranging from 25% to 0.1% with per-image no-result rates of 0% and 20%. For a PIER of 1%, the repeatability is 3 to 6 orders of magnitude higher for a 10-image FLACS system than a 2-image FLACS system.
Conclusions
FLACS biometric system repeatability can be improved by orders of magnitude by acquiring a larger number of images and applying outlier removal algorithms. Currently, commercial FLACS lasers are configured to use either 10 images or 2 images in the biometric scan. Higher biometric system repeatability may help in minimizing the risk of intra-operative complications.